FWIW, OC48 was commonplace in telco rings when I got into the business in 1996, we installed a crapload of Fujitsu FLM-2400 hardware, but mostly as transport for DS3's and the occasional STS-3 or STS-12. We certainly didn't have customers ordering whole OC48's for themselves!
But perhaps the article is blurring the notion and counting it as an OC48 if that's what was entering the building, even if a given customer only leased an OC-3 worth of it and the rest just passed through. That was certainly common, and some customers themselves were confused about the notion, so I could see the details being handwaved here as well.
As another data point, I worked on what I'm told was the first OC192 ring in Michigan, in what I believe was 1999, which included a brand new building that MCI had just constructed behind the old train station. (Fiber often follows railroads since they're straight-line easements connecting population centers.) That was the first time I encountered the notion of polarization mode dispersion compensation, which had been implemented as textbook-sized modules that slotted into the node. I was told that each module contained a bunch of fiber and some micro-actuators to stress it and try to warp the fiber into applying a complementary amount of PMD to what had happened in the OSP, restoring the pulse shape.
The Nortel TransportNode OC-192 hardware itself occupied a full rack just to break the OC192 down into OC48s, and then there were two more racks of OC48 hardware (each a half-rack) to handle customer-rate circuits. (A year or two later, when the Cerent 454 condensed a full OC48 terminal into an 8U or 6U chassis, everything changed. Cisco rapidly gobbled them up and stuck "15454" labels on them, and the rest is history.)
It likewise blows my mind that what was a whole rack of equipment then, today fits in an SFP+ that I can conceal in my palm, and a fraction of an ASIC to drive it.